The strategist's real question is which relationships could test the thesis this quarter, and what each test costs in trust. The menu below answers it item by item.
Start from a thesis that names its own counterevidence
A thesis earns a network experiment only when it predicts something about specific companies. A working version, to argue with rather than quote:
Thesis (proposed). Operating companies in our wider network hold years of records of real work that model developers cannot assemble from public text, and some of their CEOs would license those records.
Company question. Does this company keep at least five years of records it created itself, in systems it controls, with no customer claiming them?
The evidence on both sides is public, and it is more balanced than most decks admit.
| Evidence | What the source says | Effect on the thesis |
|---|---|---|
| Public text is finite | Villalobos and co-authors project that, if current trends continue, training datasets will roughly equal the stock of public human text between 2026 and 2032 | Supports scarcity of public text; says nothing about any single company's records |
| The same paper's escape routes | The authors argue that synthetic data, transfer from data-rich domains and better data efficiency might support further progress | Cuts against it: scarcity may be engineered around |
| Synthetic data in a narrow domain | phi-1, a 1.3B-parameter code model trained partly on about 1B tokens of GPT-3.5-generated textbooks and exercises, reaches 50.6% pass@1 on HumanEval | Cuts against it wherever good synthetic examples are cheap to produce |
| Program criteria | SourceX screens on size at peak, the company's own records, rights, an authorized CEO and relevance to a current program | Narrows the thesis to companies that pass, which may be few in your network |
The sources do not support a claim that data is scarce and so every company has something to sell. Our working bet is narrower: records of real operating work, with context and outcomes, are harder to generate than general text, and that bet is cheap to test with relationships the fund already holds. Market-size analysis belongs on sourcex.si.
The learning-and-introduction participation menu
Six items, ordered from inside the fund outward. The first three never approach a company. The next two reach a CEO. The last carries what was learned back to the investment team. Choose the cheapest item that could still change your mind.
| Item | What the fund contributes | What it can learn | What it must never signal | Stop rule |
|---|---|---|---|---|
| 1. Read the criteria and log questions | A platform lead's hours with the published criteria and terms, plus a dated list of open questions | Whether the fit facts match the shape of your network at all | Nothing leaves the fund | No company you know plausibly passes the size and history checks |
| 2. One metadata question in existing check-ins | One question in meetings already on the calendar: when do your own records start, and in which system? | Base rates for record age and system ownership across the network | That the fund expects an answer or will follow up | Ten answers in, nothing new comes back |
| 3. A fit review with relationship owners | One hour comparing network rows with the fit facts and recording each failed check | Which check fails most, and whether the thesis survives it | Nothing leaves the fund | The same check fails for most rows |
| 4. One permissioned introduction | A relationship owner asks one CEO; if the CEO wants it, a signed partner submits | Whether a CEO wants the conversation, and what the company chooses to share about its review | That the fund vouches for SourceX or wants the company to proceed | The CEO declines, and the row closes without a second ask |
| 5. A narrow introduction pilot | Five to eight asks, each with a prediction written down before the call | Whether the company question predicts CEO interest and fit | A campaign, or a link to anything else the fund does with the company | The result you wrote down in advance as disconfirming |
| 6. A learning memo for the investment team | A dated note of what items 1 to 5 taught about the category | Category evidence, kept apart from any view of a company or of SourceX | That a named company backs the thesis | A company would be named without its CEO's agreement |
Four things stay off the menu at every level: passing on what the fund learned from board packs or information rights (SourceX's screening guidance notes it usually stays confidential), describing a company's systems before its CEO agrees, telling a company it qualifies, and treating a referral outcome as diligence.
Platforms already give founders help that is not capital; Andreessen Horowitz, for one, describes operators in areas such as marketing, talent, legal and policy. Items 2 and 3 reuse that machinery and add one question and a log.
Method note. The menu orders items by where the work happens and marks each with a judged relationship cost; neither is measured. It applies SourceX's published fit criteria and program terms, both checked 2026-10-26. It does not estimate demand for any company's records, acceptance rates or rewards. SourceX has not published conversion data, and none is assumed here.
Who can say yes to each item
Items 1 to 3 approach no company, so the fund's own people can run them. Items 4 and 5 depend on two people the fund cannot stand in for: the relationship owner, who has agreed to ask, and the CEO, who decides whether the company will talk about licensing and is the one who signs for it. Holding shares makes the fund neither.
Before item 4, record five facts for the row, each with the date you checked it.
- The operating entity and its CEO. Who holds the records today, and who runs that entity now.
- The fit facts, marked pass, unknown or fail. Peak headcount of at least 50 full-time employees (contractors do not count), a documented operating history, rights to license, and location. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English.
- Earlier contact with SourceX. Under the terms, any earlier application, relationship, lead, account or email correspondence between SourceX and the company makes it a duplicate, and a duplicate is not credited to a later partner. Ask the CEO whether anyone at the company has ever applied, opened an account or written to SourceX.
- The corporate group. A parent and its subsidiaries share one reward cap. When a sister company has already been introduced by someone else, or the parent has dealt with SourceX before, the published terms are silent on what that means for the new introduction; SourceX settles it in duplicate review, so promise no outcome either way.
- The signed agreement. Whoever submits needs an approved partner account and a signed SourceX Channel Partner Agreement. A fund that signed the document under its old name, Channel Partner Non-Circumvention and Non-Disclosure Agreement, stays bound by that version as signed.
How to phrase the ask itself is covered in asking a business owner for permission to make an introduction.
Learning is not diligence, and an introduction is not a vote
The menu runs beside an investment process. Three rules keep the two apart.
- A referral teaches category lessons only. A company's review outcome reflects that company's records and one program's current needs. It measures no business in the category, and the terms state that partners never receive access to other partners' records, applicant documents or confidential buyer information.
- An introduction carries no fund view. Tell the CEO plainly that the fund's opinion of the company is unchanged whether it proceeds or not, and keep the subject out of board meetings.
- Declare your own interest. If the person who submits could receive a reward, say so to the CEO before the first call. The conflict questions for an investor who earns a referral reward from a portfolio company are worked through in referral fees from a company you invest in.
To keep a category bet from reading as a verdict on a company, use the attribution guide for category bets.
A six-week pilot with the losing result written first (Illustrative)
Illustrative. Every company and answer below is invented to show how the pilot reads, not to predict one.
A strategist runs item 5 on six companies, one per relationship owner, with three predictions written before any call: four of six pass the company question, two CEOs want a conversation, and no row is a duplicate.
| Company (Illustrative) | Relationship owner | Company question | CEO response | Note |
|---|---|---|---|---|
| 160-person title and escrow agency, founded 2004 | LP principal | Pass | Wants the conversation | A signed partner submits |
| 300-person regional HVAC services group | Co-investor's operating partner | Pass | Not now | Revisit on its ERP migration date |
| 70-person vertical SaaS company in the portfolio | Founder-CEO directly | Fail: product records sit in customer tenants | Not asked | Internal support records only, if any |
| 45-person analytics company founded in 2022 | Founder-CEO directly | Fail: size and history | Not asked | Revisit once full-time headcount passes 50 |
| 130-person engineering consultancy | Portfolio founder who sells to it | Pass | Declined | Closed without a second ask |
| 410-person freight brokerage | Advisor network member | Unknown: rights | Wants the conversation | Emailed SourceX last year, so a duplicate |
| Prediction written first | Result | What changes |
|---|---|---|
| At least four of six pass the company question | Three pass | The thesis narrows: portfolio software companies rarely pass, so look past the cap table |
| At least two CEOs want a conversation | Two do, one of them a duplicate | Interest appears where the company question passes |
| No duplicates | One duplicate | The earlier-contact question moves into item 2 for every row |
Two consecutive pilots in which fewer than two of six rows pass, or every passing CEO declines, would end the thesis for this network; the fund would stop at item 3 and record that in item 6. For a fuller list of findings that could sink a network thesis, see what would disprove your network's data opportunity.
Where the menu misleads
- Synthetic substitution. phi-1 shows generated examples carrying real weight in one narrow domain; a scarcity thesis should name the domains where that is least likely.
- Program demand moves. SourceX applies the criteria of a specific program and checks relevance in review. A passing company can be declined today, and one that fails a program may fit another later.
- Rights. The most interesting records may belong to a company's customers, and client-owned material is usually not a fit.
- Signal leakage. An investor's ask can feel like a request however it is worded, and no data on founder reactions has been published; log each one.
- Small numbers. Six rows cannot estimate a rate; they can only break a confident prediction.
Program basics for funds are on the venture platform teams page, and the full set of fund-side playbooks sits in the guides library.
Next step
Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.
When item 3 leaves you with a company that passes every check you can confirm and a relationship owner willing to ask, the pilot's first introduction is in reach: register for a partner link, run the row past who qualifies one last time and go over the program terms before the ask is made.
Choosing which relationships to test is referral work. Deciding whether SourceX itself merits an investment is a separate exercise; for that, request investor materials through the SourceX investor overview.
This article is for information only and is not an offer to sell or a solicitation of an offer to buy securities. Any offering by SourceX is made only to verified accredited investors through its offering documents.